
# Загрузка библиотек, необходимых в проекте:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import re

from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import mean_absolute_error, roc_auc_score, accuracy_score, confusion_matrix
from sklearn.model_selection import cross_val_score
from sklearn.preprocessing import FunctionTransformer, OneHotEncoder, OrdinalEncoder, StandardScaler
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_val_score, RandomizedSearchCV, GridSearchCV, train_test_split
from sklearn.metrics import roc_curve 
from sklearn.pipeline import Pipeline, FeatureUnion
from sklearn.compose import ColumnTransformer


!pip install phik
import phik
from phik import resources
from phik.binning import bin_data
from phik.report import plot_correlation_matrix


!pip install catboost
import catboost as cat
from catboost import CatBoostClassifier, Pool, cv

import warnings
warnings.filterwarnings("ignore")

!pip install lightgbm
import lightgbm
from lightgbm import LGBMClassifier

Requirement already satisfied: phik in c:\users\user\anaconda3\lib\site-packages (0.12.3)
Requirement already satisfied: numpy>=1.18.0 in c:\users\user\anaconda3\lib\site-packages (from phik) (1.24.3)
Requirement already satisfied: scipy>=1.5.2 in c:\users\user\anaconda3\lib\site-packages (from phik) (1.10.1)
Requirement already satisfied: pandas>=0.25.1 in c:\users\user\anaconda3\lib\site-packages (from phik) (1.5.3)
Requirement already satisfied: matplotlib>=2.2.3 in c:\users\user\anaconda3\lib\site-packages (from phik) (3.7.1)
Requirement already satisfied: joblib>=0.14.1 in c:\users\user\anaconda3\lib\site-packages (from phik) (1.2.0)
Requirement already satisfied: contourpy>=1.0.1 in c:\users\user\anaconda3\lib\site-packages (from matplotlib>=2.2.3->phik) (1.0.5)
Requirement already satisfied: cycler>=0.10 in c:\users\user\anaconda3\lib\site-packages (from matplotlib>=2.2.3->phik) (0.11.0)
Requirement already satisfied: fonttools>=4.22.0 in c:\users\user\anaconda3\lib\site-packages (from matplotlib>=2.2.3->phik) (4.25.0)
Requirement already satisfied: kiwisolver>=1.0.1 in c:\users\user\anaconda3\lib\site-packages (from matplotlib>=2.2.3->phik) (1.4.4)
Requirement already satisfied: packaging>=20.0 in c:\users\user\anaconda3\lib\site-packages (from matplotlib>=2.2.3->phik) (23.0)
Requirement already satisfied: pillow>=6.2.0 in c:\users\user\anaconda3\lib\site-packages (from matplotlib>=2.2.3->phik) (9.4.0)
Requirement already satisfied: pyparsing>=2.3.1 in c:\users\user\anaconda3\lib\site-packages (from matplotlib>=2.2.3->phik) (3.0.9)
Requirement already satisfied: python-dateutil>=2.7 in c:\users\user\anaconda3\lib\site-packages (from matplotlib>=2.2.3->phik) (2.8.2)
Requirement already satisfied: importlib-resources>=3.2.0 in c:\users\user\anaconda3\lib\site-packages (from matplotlib>=2.2.3->phik) (5.2.0)
Requirement already satisfied: pytz>=2020.1 in c:\users\user\anaconda3\lib\site-packages (from pandas>=0.25.1->phik) (2022.7)
Requirement already satisfied: zipp>=3.1.0 in c:\users\user\anaconda3\lib\site-packages (from importlib-resources>=3.2.0->matplotlib>=2.2.3->phik) (3.11.0)
Requirement already satisfied: six>=1.5 in c:\users\user\anaconda3\lib\site-packages (from python-dateutil>=2.7->matplotlib>=2.2.3->phik) (1.16.0)
Requirement already satisfied: catboost in c:\users\user\anaconda3\lib\site-packages (1.2)
Requirement already satisfied: graphviz in c:\users\user\anaconda3\lib\site-packages (from catboost) (0.20.1)
Requirement already satisfied: matplotlib in c:\users\user\anaconda3\lib\site-packages (from catboost) (3.7.1)
Requirement already satisfied: numpy>=1.16.0 in c:\users\user\anaconda3\lib\site-packages (from catboost) (1.24.3)
Requirement already satisfied: pandas>=0.24 in c:\users\user\anaconda3\lib\site-packages (from catboost) (1.5.3)
Requirement already satisfied: scipy in c:\users\user\anaconda3\lib\site-packages (from catboost) (1.10.1)
Requirement already satisfied: plotly in c:\users\user\anaconda3\lib\site-packages (from catboost) (5.9.0)
Requirement already satisfied: six in c:\users\user\anaconda3\lib\site-packages (from catboost) (1.16.0)
Requirement already satisfied: python-dateutil>=2.8.1 in c:\users\user\anaconda3\lib\site-packages (from pandas>=0.24->catboost) (2.8.2)
Requirement already satisfied: pytz>=2020.1 in c:\users\user\anaconda3\lib\site-packages (from pandas>=0.24->catboost) (2022.7)
Requirement already satisfied: contourpy>=1.0.1 in c:\users\user\anaconda3\lib\site-packages (from matplotlib->catboost) (1.0.5)
Requirement already satisfied: cycler>=0.10 in c:\users\user\anaconda3\lib\site-packages (from matplotlib->catboost) (0.11.0)
Requirement already satisfied: fonttools>=4.22.0 in c:\users\user\anaconda3\lib\site-packages (from matplotlib->catboost) (4.25.0)
Requirement already satisfied: kiwisolver>=1.0.1 in c:\users\user\anaconda3\lib\site-packages (from matplotlib->catboost) (1.4.4)
Requirement already satisfied: packaging>=20.0 in c:\users\user\anaconda3\lib\site-packages (from matplotlib->catboost) (23.0)
Requirement already satisfied: pillow>=6.2.0 in c:\users\user\anaconda3\lib\site-packages (from matplotlib->catboost) (9.4.0)
Requirement already satisfied: pyparsing>=2.3.1 in c:\users\user\anaconda3\lib\site-packages (from matplotlib->catboost) (3.0.9)
Requirement already satisfied: importlib-resources>=3.2.0 in c:\users\user\anaconda3\lib\site-packages (from matplotlib->catboost) (5.2.0)
Requirement already satisfied: tenacity>=6.2.0 in c:\users\user\anaconda3\lib\site-packages (from plotly->catboost) (8.2.2)
Requirement already satisfied: zipp>=3.1.0 in c:\users\user\anaconda3\lib\site-packages (from importlib-resources>=3.2.0->matplotlib->catboost) (3.11.0)
Requirement already satisfied: lightgbm in c:\users\user\anaconda3\lib\site-packages (3.3.5)
Requirement already satisfied: wheel in c:\users\user\anaconda3\lib\site-packages (from lightgbm) (0.38.4)
Requirement already satisfied: numpy in c:\users\user\anaconda3\lib\site-packages (from lightgbm) (1.24.3)
Requirement already satisfied: scipy in c:\users\user\anaconda3\lib\site-packages (from lightgbm) (1.10.1)
Requirement already satisfied: scikit-learn!=0.22.0 in c:\users\user\anaconda3\lib\site-packages (from lightgbm) (1.3.0)
Requirement already satisfied: joblib>=1.1.1 in c:\users\user\anaconda3\lib\site-packages (from scikit-learn!=0.22.0->lightgbm) (1.2.0)
Requirement already satisfied: threadpoolctl>=2.0.0 in c:\users\user\anaconda3\lib\site-packages (from scikit-learn!=0.22.0->lightgbm) (2.2.0)


!pip install -U scikit-learn

Requirement already satisfied: scikit-learn in c:\users\user\anaconda3\lib\site-packages (1.3.0)
Requirement already satisfied: numpy>=1.17.3 in c:\users\user\anaconda3\lib\site-packages (from scikit-learn) (1.24.3)
Requirement already satisfied: scipy>=1.5.0 in c:\users\user\anaconda3\lib\site-packages (from scikit-learn) (1.10.1)
Requirement already satisfied: joblib>=1.1.1 in c:\users\user\anaconda3\lib\site-packages (from scikit-learn) (1.2.0)
Requirement already satisfied: threadpoolctl>=2.0.0 in c:\users\user\anaconda3\lib\site-packages (from scikit-learn) (2.2.0)


# Загрузим данные из файла:

# contract_new
try:
    contract_new = pd.read_csv('') 
except:
    try:
        contract_new = pd.read_csv('D:/YandexDisk/Яндекс практикум/18 Финальный проект/contract_new.csv') # локально Windiws work
    except:
        contract_new = pd.read_csv('/datasets/contract_new.csv') # онлайн
contract_new.name = 'Информация о договоре'  
        
# personal_new
try:
    personal_new = pd.read_csv('') # онлайн
except:
    try:
        personal_new = pd.read_csv('D:/YandexDisk/Яндекс практикум/18 Финальный проект/personal_new.csv') # локально Windiws work
    except:
        personal_new = pd.read_csv('/datasets/personal_new.csv') # онлайн
personal_new.name = 'Персональные данные клиента' 

        
# internet_new
try:
    internet_new = pd.read_csv('') # онлайн
except:
    try:
        internet_new = pd.read_csv('D:/YandexDisk/Яндекс практикум/18 Финальный проект/internet_new.csv') # локально Windiws work
    except:
        internet_new = pd.read_csv('/datasets/internet_new.csv') # онлайн
internet_new.name = 'Информация об интернет-услугах' 
                         

# phone_new
try:
    phone_new = pd.read_csv('https://code.s3.yandex.net/datasets/phone_new.csv') # онлайн
except:
    try:
        phone_new = pd.read_csv('D:/YandexDisk/Яндекс практикум/18 Финальный проект/phone_new.csv') # локально Windiws work
    except:
        phone_new = pd.read_csv('/datasets/phone_new.csv') # онлайн
phone_new.name = 'Информация об услугах телефонии'


def data_analysis (data):
    print ('\033[1m' + data.name +'\033[0m')
    display(data.head()),
    print()
    print('Количество строк:', data.shape[0])
    print('Количество столбцов:', data.shape[1])
    print()
    print('Информация о таблице:')
    data.info()
    print()
    print()


data_list = [contract_new, personal_new, internet_new, phone_new]
for i in data_list:
    data_analysis(i)

Информация о договоре

Количество строк: 7043
Количество столбцов: 8

Информация о таблице:
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 7043 entries, 0 to 7042
Data columns (total 8 columns):
 #   Column            Non-Null Count  Dtype  
---  ------            --------------  -----  
 0   customerID        7043 non-null   object 
 1   BeginDate         7043 non-null   object 
 2   EndDate           7043 non-null   object 
 3   Type              7043 non-null   object 
 4   PaperlessBilling  7043 non-null   object 
 5   PaymentMethod     7043 non-null   object 
 6   MonthlyCharges    7043 non-null   float64
 7   TotalCharges      7043 non-null   object 
dtypes: float64(1), object(7)
memory usage: 440.3+ KB


Персональные данные клиента

Количество строк: 7043
Количество столбцов: 5

Информация о таблице:
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 7043 entries, 0 to 7042
Data columns (total 5 columns):
 #   Column         Non-Null Count  Dtype 
---  ------         --------------  ----- 
 0   customerID     7043 non-null   object
 1   gender         7043 non-null   object
 2   SeniorCitizen  7043 non-null   int64 
 3   Partner        7043 non-null   object
 4   Dependents     7043 non-null   object
dtypes: int64(1), object(4)
memory usage: 275.2+ KB


Информация об интернет-услугах

Количество строк: 5517
Количество столбцов: 8

Информация о таблице:
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 5517 entries, 0 to 5516
Data columns (total 8 columns):
 #   Column            Non-Null Count  Dtype 
---  ------            --------------  ----- 
 0   customerID        5517 non-null   object
 1   InternetService   5517 non-null   object
 2   OnlineSecurity    5517 non-null   object
 3   OnlineBackup      5517 non-null   object
 4   DeviceProtection  5517 non-null   object
 5   TechSupport       5517 non-null   object
 6   StreamingTV       5517 non-null   object
 7   StreamingMovies   5517 non-null   object
dtypes: object(8)
memory usage: 344.9+ KB


Информация об услугах телефонии

Количество строк: 6361
Количество столбцов: 2

Информация о таблице:
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 6361 entries, 0 to 6360
Data columns (total 2 columns):
 #   Column         Non-Null Count  Dtype 
---  ------         --------------  ----- 
 0   customerID     6361 non-null   object
 1   MultipleLines  6361 non-null   object
dtypes: object(2)
memory usage: 99.5+ KB


df = contract_new.merge(
    personal_new, on='customerID', how='outer'
).merge(
    internet_new, on='customerID', how='outer'
).merge(
    phone_new, on='customerID', how='outer'
)
df.name = 'Объединенная таблица'


print('Количество строк:', df.shape[0])
print('Количество столбцов:', df.shape[1])
print('Название столбцов:', df.columns.values)

Количество строк: 7043
Количество столбцов: 20
Название столбцов: ['customerID' 'BeginDate' 'EndDate' 'Type' 'PaperlessBilling'
 'PaymentMethod' 'MonthlyCharges' 'TotalCharges' 'gender' 'SeniorCitizen'
 'Partner' 'Dependents' 'InternetService' 'OnlineSecurity' 'OnlineBackup'
 'DeviceProtection' 'TechSupport' 'StreamingTV' 'StreamingMovies'
 'MultipleLines']


# код тимлида для проверки
data_analysis(df)

Объединенная таблица

Количество строк: 7043
Количество столбцов: 20

Информация о таблице:
<class 'pandas.core.frame.DataFrame'>
Int64Index: 7043 entries, 0 to 7042
Data columns (total 20 columns):
 #   Column            Non-Null Count  Dtype  
---  ------            --------------  -----  
 0   customerID        7043 non-null   object 
 1   BeginDate         7043 non-null   object 
 2   EndDate           7043 non-null   object 
 3   Type              7043 non-null   object 
 4   PaperlessBilling  7043 non-null   object 
 5   PaymentMethod     7043 non-null   object 
 6   MonthlyCharges    7043 non-null   float64
 7   TotalCharges      7043 non-null   object 
 8   gender            7043 non-null   object 
 9   SeniorCitizen     7043 non-null   int64  
 10  Partner           7043 non-null   object 
 11  Dependents        7043 non-null   object 
 12  InternetService   5517 non-null   object 
 13  OnlineSecurity    5517 non-null   object 
 14  OnlineBackup      5517 non-null   object 
 15  DeviceProtection  5517 non-null   object 
 16  TechSupport       5517 non-null   object 
 17  StreamingTV       5517 non-null   object 
 18  StreamingMovies   5517 non-null   object 
 19  MultipleLines     6361 non-null   object 
dtypes: float64(1), int64(1), object(18)
memory usage: 1.1+ MB


def convert_to_snake(columns):
    return columns.str.replace('(?<=[a-z])(?=[A-Z])', '_', regex=True).str.lower()


df.columns = convert_to_snake(df.columns)
df.columns

Index(['customer_id', 'begin_date', 'end_date', 'type', 'paperless_billing',
       'payment_method', 'monthly_charges', 'total_charges', 'gender',
       'senior_citizen', 'partner', 'dependents', 'internet_service',
       'online_security', 'online_backup', 'device_protection', 'tech_support',
       'streaming_tv', 'streaming_movies', 'multiple_lines'],
      dtype='object')


df['end_date'].unique()

array(['No', '2017-05-01', '2016-03-01', '2018-09-01', '2018-11-01',
       '2018-12-01', '2019-08-01', '2018-07-01', '2017-09-01',
       '2015-09-01', '2016-07-01', '2016-06-01', '2018-03-01',
       '2019-02-01', '2018-06-01', '2019-06-01', '2020-01-01',
       '2019-11-01', '2016-09-01', '2015-06-01', '2016-12-01',
       '2019-05-01', '2019-04-01', '2017-06-01', '2017-08-01',
       '2018-04-01', '2018-08-01', '2018-02-01', '2019-07-01',
       '2015-12-01', '2014-06-01', '2018-10-01', '2019-01-01',
       '2017-07-01', '2017-12-01', '2018-05-01', '2015-11-01',
       '2019-10-01', '2019-03-01', '2016-02-01', '2016-10-01',
       '2018-01-01', '2017-11-01', '2015-10-01', '2019-12-01',
       '2015-07-01', '2017-04-01', '2015-02-01', '2017-03-01',
       '2016-05-01', '2016-11-01', '2015-08-01', '2019-09-01',
       '2017-10-01', '2017-02-01', '2016-08-01', '2016-04-01',
       '2015-05-01', '2014-09-01', '2014-10-01', '2017-01-01',
       '2015-03-01', '2015-01-01', '2016-01-01', '2015-04-01',
       '2014-12-01', '2014-11-01'], dtype=object)


df['leave'] = (
    df['end_date'].where(
        df['end_date'] == 'No', 'Yes')
)


display(df[df['end_date'] != 'No'][['begin_date','end_date','leave']].head())
df[df['end_date'] == 'No'][['begin_date','end_date','leave']].head()


print ('Процентное соотношение значений столбца "leave":\n', 
       round(df['leave'].value_counts()/df['leave'].count()*100))
print ()

(df['leave'].value_counts()/df['leave'].count()*100).plot(
    kind='bar', 
    grid = True
).set(
    ylabel = 'Относительное количество, %', 
    xlabel = 'Факт ухода клиента: \n "No" - Клиент не ушел, "Yes" - Клиент ушел', 
    title = 'Распределение оставшихся и ушедщих клиентов' 
)
plt.show()

Процентное соотношение значений столбца "leave":
 No     84.0
Yes    16.0
Name: leave, dtype: float64


# код тимлида для проверки
df['leave'].value_counts(normalize=True)

No     0.843675
Yes    0.156325
Name: leave, dtype: float64


pd.set_option('display.max_columns', None)
data_analysis(df)
print()
print('Количество дупликатов:', df['customer_id'].duplicated().sum())
print('Количество пропусков:', df.isna().sum().sum())
print('Количество строк с пропусками:',  df[df.isna().sum(axis= 1) > 0].shape[0])
print()

Объединенная таблица

Количество строк: 7043
Количество столбцов: 21

Информация о таблице:
<class 'pandas.core.frame.DataFrame'>
Int64Index: 7043 entries, 0 to 7042
Data columns (total 21 columns):
 #   Column             Non-Null Count  Dtype  
---  ------             --------------  -----  
 0   customer_id        7043 non-null   object 
 1   begin_date         7043 non-null   object 
 2   end_date           7043 non-null   object 
 3   type               7043 non-null   object 
 4   paperless_billing  7043 non-null   object 
 5   payment_method     7043 non-null   object 
 6   monthly_charges    7043 non-null   float64
 7   total_charges      7043 non-null   object 
 8   gender             7043 non-null   object 
 9   senior_citizen     7043 non-null   int64  
 10  partner            7043 non-null   object 
 11  dependents         7043 non-null   object 
 12  internet_service   5517 non-null   object 
 13  online_security    5517 non-null   object 
 14  online_backup      5517 non-null   object 
 15  device_protection  5517 non-null   object 
 16  tech_support       5517 non-null   object 
 17  streaming_tv       5517 non-null   object 
 18  streaming_movies   5517 non-null   object 
 19  multiple_lines     6361 non-null   object 
 20  leave              7043 non-null   object 
dtypes: float64(1), int64(1), object(19)
memory usage: 1.2+ MB


Количество дупликатов: 0
Количество пропусков: 11364
Количество строк с пропусками: 2208


print('Количество пенсионеров до преобразования:', df[df['senior_citizen'] == 1]['senior_citizen'].count(), ', Тип данных:', df['senior_citizen'].dtype)

df['senior_citizen'] = (
    df['senior_citizen'].mask(
        df['senior_citizen'] == 0, 'No').where(
        df['senior_citizen'] == 0, 'Yes')
)
print('Количество пенсионеров после преобразования:', df[df['senior_citizen'] == 'Yes']['senior_citizen'].count(), ', Тип данных:', df['senior_citizen'].dtype)

Количество пенсионеров до преобразования: 1142 , Тип данных: int64
Количество пенсионеров после преобразования: 1142 , Тип данных: object


print('Тип данных столбца "total_charges" до преобразования:', df['total_charges'].dtype)
df['total_charges'] = pd.to_numeric(df['total_charges'],errors='coerce')
print('Тип данных столбца "total_charges" после преобразования:', df['total_charges'].dtype)

Тип данных столбца "total_charges" до преобразования: object
Тип данных столбца "total_charges" после преобразования: float64


print('Тип данных столбцов до преобразования:')
print('begin_date -', df['begin_date'].dtype)
print('end_date -', df['end_date'].dtype)

df['begin_date'] = pd.to_datetime(df['begin_date'], errors='ignore')
df['end_date'] = pd.to_datetime(df['end_date'], errors='coerce')

print('Тип данных столбцов после преобразования:')
print('begin_date -', df['begin_date'].dtype)
print('end_date -', df['end_date'].dtype)

Тип данных столбцов до преобразования:
begin_date - object
end_date - object
Тип данных столбцов после преобразования:
begin_date - datetime64[ns]
end_date - datetime64[ns]


df[df['total_charges'].isna()== True].sample(5)


print('Количество пропусков до заполнения:', df.total_charges.isna().sum())

df['total_charges'] = df['total_charges'].fillna(0)

print('Количество пропусков после заполнения:', df.total_charges.isna().sum())

Количество пропусков до заполнения: 11
Количество пропусков после заполнения: 0


print('Количество пропусков до заполнения:', df.end_date.isna().sum())

df['end_date'] = df['end_date'].fillna('2020-02-01')

print('Количество пропусков после заполнения:', df.end_date.isna().sum())

Количество пропусков до заполнения: 5942
Количество пропусков после заполнения: 0


print('Заполненных строк:', df.multiple_lines.isna().sum())
print()

print ('\033[1m' + 'До заполнения:' +'\033[0m')
print('Уникальные значения:', df['multiple_lines'].unique())
print('Количество строк с пропусками:', df[df.isna().sum(axis= 1) > 0].shape[0])
    
df['multiple_lines'] = df['multiple_lines'].fillna('No service')

print()
print ('\033[1m' + 'После заполнения:' +'\033[0m')
print('Уникальные значения:', df['multiple_lines'].unique())
print('Количество строк с пропусками:', df[df.isna().sum(axis= 1) > 0].shape[0])

Заполненных строк: 682

До заполнения:
Уникальные значения: [nan 'No' 'Yes']
Количество строк с пропусками: 2208

После заполнения:
Уникальные значения: ['No service' 'No' 'Yes']
Количество строк с пропусками: 1526


print('Заполненных строк:', df.internet_service.isna().sum())
print()

print ('\033[1m' + 'До заполнения:' +'\033[0m')
print('Количество строк с пропусками:', df[df.isna().sum(axis= 1) > 0].shape[0])
print('Уникальные значения:')
for i in ['internet_service', 'online_security', 'online_backup', 
          'device_protection', 'tech_support', 'streaming_tv', 'streaming_movies']:
    print(i,'-', df[i].unique())

# Заполнение 'online_security', 'online_backup', 'device_protection', 'tech_support', 'streaming_tv', 'streaming_movies'
# при условии, что в 'internet_service' есть пропуски :
df.loc[df['internet_service'].isna() == True, 
       ['online_security','online_backup', 'device_protection', 'tech_support', 'streaming_tv', 'streaming_movies']
      ] = df.loc[df['internet_service'].isna() == True, 
                 ['online_security','online_backup', 'device_protection', 'tech_support', 'streaming_tv', 'streaming_movies']
                ].fillna('No')

# Заполнение 'internet_service':
df['internet_service'] = df['internet_service'].fillna('No service')

print()
print ('\033[1m' + 'После заполнения:' +'\033[0m')
print('Количество строк с пропусками:', df[df.isna().sum(axis= 1) > 0].shape[0])
print('Уникальные значения:')
for i in ['internet_service', 'online_security', 'online_backup', 
          'device_protection', 'tech_support', 'streaming_tv', 'streaming_movies']:
    print(i,'-', df[i].unique())

Заполненных строк: 1526

До заполнения:
Количество строк с пропусками: 1526
Уникальные значения:
internet_service - ['DSL' 'Fiber optic' nan]
online_security - ['No' 'Yes' nan]
online_backup - ['Yes' 'No' nan]
device_protection - ['No' 'Yes' nan]
tech_support - ['No' 'Yes' nan]
streaming_tv - ['No' 'Yes' nan]
streaming_movies - ['No' 'Yes' nan]

После заполнения:
Количество строк с пропусками: 0
Уникальные значения:
internet_service - ['DSL' 'Fiber optic' 'No service']
online_security - ['No' 'Yes']
online_backup - ['Yes' 'No']
device_protection - ['No' 'Yes']
tech_support - ['No' 'Yes']
streaming_tv - ['No' 'Yes']
streaming_movies - ['No' 'Yes']


data_analysis(df)

Объединенная таблица

Количество строк: 7043
Количество столбцов: 21

Информация о таблице:
<class 'pandas.core.frame.DataFrame'>
Int64Index: 7043 entries, 0 to 7042
Data columns (total 21 columns):
 #   Column             Non-Null Count  Dtype         
---  ------             --------------  -----         
 0   customer_id        7043 non-null   object        
 1   begin_date         7043 non-null   datetime64[ns]
 2   end_date           7043 non-null   datetime64[ns]
 3   type               7043 non-null   object        
 4   paperless_billing  7043 non-null   object        
 5   payment_method     7043 non-null   object        
 6   monthly_charges    7043 non-null   float64       
 7   total_charges      7043 non-null   float64       
 8   gender             7043 non-null   object        
 9   senior_citizen     7043 non-null   object        
 10  partner            7043 non-null   object        
 11  dependents         7043 non-null   object        
 12  internet_service   7043 non-null   object        
 13  online_security    7043 non-null   object        
 14  online_backup      7043 non-null   object        
 15  device_protection  7043 non-null   object        
 16  tech_support       7043 non-null   object        
 17  streaming_tv       7043 non-null   object        
 18  streaming_movies   7043 non-null   object        
 19  multiple_lines     7043 non-null   object        
 20  leave              7043 non-null   object        
dtypes: datetime64[ns](2), float64(2), object(17)
memory usage: 1.2+ MB


def analysis_quantitatives(features_list):
    for i in features_list:
        print()
        print('\033[1m' + i,':' +'\033[0m')
        print()
        print('Клиенты ушли от оператора:')
        print(df[i][df['leave'] == 'Yes'].describe())
        print()
        print('Клиенты не ушли от оператора:')
        print(df[i][df['leave'] == 'No'].describe())
        sns.set()
        lines, axes = plt.subplots(1, 2, figsize=(16, 4))
        sns.histplot(data=df, hue='leave', x=i, ax=axes[0], kde=True).set_title(i + ". Плотность распределения", fontsize=18)
        sns.boxplot(data=df, x=i, y='leave', ax=axes[1]).set_title(i + ". Диаграмма размаха", fontsize=18)
        plt.show()


analysis_quantitatives(['monthly_charges','total_charges'])

monthly_charges :

Клиенты ушли от оператора:
count    1101.000000
mean       75.546004
std        29.116213
min        18.400000
25%        56.150000
50%        84.200000
75%        99.500000
max       118.750000
Name: monthly_charges, dtype: float64

Клиенты не ушли от оператора:
count    5942.000000
mean       62.763455
std        29.844462
min        18.250000
25%        30.062500
50%        69.200000
75%        87.237500
max       118.600000
Name: monthly_charges, dtype: float64

total_charges :

Клиенты ушли от оператора:
count    1101.000000
mean     2371.377275
std      1581.862275
min        77.840000
25%      1048.050000
50%      2139.030000
75%      3404.910000
max      7649.760000
Name: total_charges, dtype: float64

Клиенты не ушли от оператора:
count    5942.000000
mean     2067.866420
std      2193.898483
min         0.000000
25%       374.352500
50%      1192.800000
75%      3173.837500
max      9221.380000
Name: total_charges, dtype: float64


df[df['total_charges']>7500].sample(10)


col_list = ['type', 'paperless_billing', 'gender', 'senior_citizen',
            'partner', 'dependents', 'payment_method', 'internet_service', 'online_security', 'online_backup',
            'device_protection', 'tech_support', 'streaming_tv', 'streaming_movies', 'multiple_lines']
name_list = ['Тип договора', 'Выставления счёта по электронной почте', 'Пол клиента', 'Пенсионный статус',
            'Наличие супруга/супруги', 'Наличие иждивенцев', 'Способ оплаты', 'Наличие услуг Интернет', 'Межсетевой экран', 
             'Облачное хранилище файлов', 'Антивирус', 'Выделенная линия технической поддержки', 
             'Онлайн-ТВ', 'Онлайн-кинотеатр', 'Возможность подключения телефонного \n аппарата к нескольким линиям']


pic_box = plt.figure(figsize=(16,20))
 
for i in range(15):
    pic_box.add_subplot(5,3,i+1)
    sns.histplot(data=df, x=df[col_list[i]], hue='leave', multiple='dodge', shrink = 0.8)
    plt.title(str(i+1)+'.'+ name_list[i], fontsize=14)
plt.tight_layout()


for i in range(15):
    plt.figure(figsize=(12,4))
    for b in df[col_list[i]].unique(): 
        plt.subplot(1,len(df[col_list[i]].unique()),list(df[col_list[i]].unique()).index(b)+1)
        sns.histplot(data=df[df[col_list[i]] == b], x='leave', hue ='leave',
                     shrink = 0.8, multiple='dodge', stat = 'percent', bins = 4)
        plt.title(str(i+1)+'.'+ name_list[i] + '\n- ' + b, fontsize=12)
plt.tight_layout()


# Длительность обслуживания клиента 'duration'
df['duration'] = (df['end_date'] - df['begin_date'])/np.timedelta64(1,'D')


def contract_end(data):
    if data['type'] == 'Month-to-month':
        return (data['end_date'] - data['begin_date'])/np.timedelta64(1,'M')\
        - np.floor((data['end_date'] - data['begin_date'])/np.timedelta64(1,'M')) 
    if data['type'] == 'One year':
        return (data['end_date'] - data['begin_date'])/np.timedelta64(1,'Y')\
        - np.floor((data['end_date'] - data['begin_date'])/np.timedelta64(1,'Y')) 
    if data['type'] == 'Two year':
        return (data['end_date'] - data['begin_date'])/np.timedelta64(2,'Y')\
        - np.floor((data['end_date'] - data['begin_date'])/np.timedelta64(2,'Y'))


df['contract_end'] = df.apply(contract_end, axis = 1)


analysis_quantitatives(['duration', 'contract_end'])

duration :

Клиенты ушли от оператора:
count    1101.000000
mean      924.863760
std       458.771309
min        28.000000
25%       577.000000
50%       915.000000
75%      1249.000000
max      2129.000000
Name: duration, dtype: float64

Клиенты не ушли от оператора:
count    5942.000000
mean      893.681084
std       716.958551
min         0.000000
25%       245.000000
50%       702.000000
75%      1523.000000
max      2314.000000
Name: duration, dtype: float64

contract_end :

Клиенты ушли от оператора:
count    1101.000000
mean        0.483135
std         0.364089
min         0.000041
25%         0.085621
50%         0.457935
75%         0.835767
max         0.999418
Name: contract_end, dtype: float64

Клиенты не ушли от оператора:
count    5942.000000
mean        0.310056
std         0.367864
min         0.000000
25%         0.033183
50%         0.063800
75%         0.625653
max         0.999418
Name: contract_end, dtype: float64


df_before_delete = df.copy()
    
df.drop(['customer_id', 'begin_date', 'end_date'], axis=1, inplace = True)


data_analysis(df)

Объединенная таблица

Количество строк: 7043
Количество столбцов: 20

Информация о таблице:
<class 'pandas.core.frame.DataFrame'>
Int64Index: 7043 entries, 0 to 7042
Data columns (total 20 columns):
 #   Column             Non-Null Count  Dtype  
---  ------             --------------  -----  
 0   type               7043 non-null   object 
 1   paperless_billing  7043 non-null   object 
 2   payment_method     7043 non-null   object 
 3   monthly_charges    7043 non-null   float64
 4   total_charges      7043 non-null   float64
 5   gender             7043 non-null   object 
 6   senior_citizen     7043 non-null   object 
 7   partner            7043 non-null   object 
 8   dependents         7043 non-null   object 
 9   internet_service   7043 non-null   object 
 10  online_security    7043 non-null   object 
 11  online_backup      7043 non-null   object 
 12  device_protection  7043 non-null   object 
 13  tech_support       7043 non-null   object 
 14  streaming_tv       7043 non-null   object 
 15  streaming_movies   7043 non-null   object 
 16  multiple_lines     7043 non-null   object 
 17  leave              7043 non-null   object 
 18  duration           7043 non-null   float64
 19  contract_end       7043 non-null   float64
dtypes: float64(4), object(16)
memory usage: 1.1+ MB


phik_overview = df[['leave'] + [x for x in df.columns if x != 'leave']
                  ].phik_matrix(interval_cols=['monthly_charges','total_charges', 'duration', 'contract_end'])


phik_overview.sort_values(by='leave', ascending=False).style.background_gradient(cmap = 'coolwarm')


plot_correlation_matrix(phik_overview.values, x_labels=phik_overview.columns, y_labels=phik_overview.index, 
                        vmin=0, vmax=1, color_map='Blues', title=r'correlation $\phi_K$', fontsize_factor=1.5,
                        figsize=(16,20))
plt.tight_layout()


 df.drop(['gender', 'dependents', 'internet_service'], axis=1, inplace = True)


df.columns

Index(['type', 'paperless_billing', 'payment_method', 'monthly_charges',
       'total_charges', 'senior_citizen', 'partner', 'online_security',
       'online_backup', 'device_protection', 'tech_support', 'streaming_tv',
       'streaming_movies', 'multiple_lines', 'leave', 'duration',
       'contract_end'],
      dtype='object')


data_analysis(df)

Объединенная таблица

Количество строк: 7043
Количество столбцов: 17

Информация о таблице:
<class 'pandas.core.frame.DataFrame'>
Int64Index: 7043 entries, 0 to 7042
Data columns (total 17 columns):
 #   Column             Non-Null Count  Dtype  
---  ------             --------------  -----  
 0   type               7043 non-null   object 
 1   paperless_billing  7043 non-null   object 
 2   payment_method     7043 non-null   object 
 3   monthly_charges    7043 non-null   float64
 4   total_charges      7043 non-null   float64
 5   senior_citizen     7043 non-null   object 
 6   partner            7043 non-null   object 
 7   online_security    7043 non-null   object 
 8   online_backup      7043 non-null   object 
 9   device_protection  7043 non-null   object 
 10  tech_support       7043 non-null   object 
 11  streaming_tv       7043 non-null   object 
 12  streaming_movies   7043 non-null   object 
 13  multiple_lines     7043 non-null   object 
 14  leave              7043 non-null   object 
 15  duration           7043 non-null   float64
 16  contract_end       7043 non-null   float64
dtypes: float64(4), object(13)
memory usage: 990.4+ KB


target = df['leave']
features = df.drop(['leave'], axis=1)


print('Количество уникальных значений до замены:')
print(target.value_counts())

target = target.replace(['No', 'Yes'], [0, 1])

print()
print('Количество уникальных значений после замены:')
print(target.value_counts())

Количество уникальных значений до замены:
No     5942
Yes    1101
Name: leave, dtype: int64

Количество уникальных значений после замены:
0    5942
1    1101
Name: leave, dtype: int64


features_train, features_test, target_train, target_test = train_test_split(
    features, target, test_size=0.33, random_state=140823, stratify=target, shuffle = True)

print('"features_train":', features_train.shape[0],'объектов,', 
      100 - round((features.shape[0] - features_train.shape[0]) * 100 / features.shape[0]), "%" )
print('"target_train":', target_train.shape[0],'объектов,', 
      100 - round((target.shape[0] - target_train.shape[0]) * 100 / target.shape[0]), "%" )
print('"features_test":', features_test.shape[0],'объектов,', 
      100 - round((features.shape[0] - features_test.shape[0]) * 100 / features.shape[0]), "%" )
print('"target_test":', target_test.shape[0],'объектов,', 
      100 - round((target.shape[0] - target_test.shape[0]) * 100 / target.shape[0]), "%" )

"features_train": 4718 объектов, 67 %
"target_train": 4718 объектов, 67 %
"features_test": 2325 объектов, 33 %
"target_test": 2325 объектов, 33 %


numerical = features_train.select_dtypes(include=[np.number]).columns.tolist()
print('Числовые признаки:', numerical)
print()
categorical = features_train.select_dtypes(include='object').columns.tolist()
print('Категориальные признаки:', categorical)

Числовые признаки: ['monthly_charges', 'total_charges', 'duration', 'contract_end']

Категориальные признаки: ['type', 'paperless_billing', 'payment_method', 'senior_citizen', 'partner', 'online_security', 'online_backup', 'device_protection', 'tech_support', 'streaming_tv', 'streaming_movies', 'multiple_lines']


numerical_transformer = StandardScaler() 
categorical_transformer_ohe = OneHotEncoder(sparse=False, handle_unknown='ignore')
categorical_transformer_ord_encoder = OrdinalEncoder()

# Делаем ColumnTransformer для категориальных переменных с OrdinalEncoder
preprocessor_ohe = ColumnTransformer(
    transformers=[
        ('num', numerical_transformer, numerical),
        ('cat', categorical_transformer_ohe, categorical)], verbose_feature_names_out=True)

# Делаем ColumnTransformer для категориальных переменных с OrdinalEncoder
preprocessor_ord_encoder = ColumnTransformer(
    transformers=[
        ('num', numerical_transformer, numerical),
        ('cat', categorical_transformer_ord_encoder, categorical)], verbose_feature_names_out=True)


%%time

#  Составляем pipeline:

pipeline_LR = Pipeline([
    ("preprocessor_ohe", preprocessor_ohe),
    ('model_LR', LogisticRegression(random_state=140823,
                                    penalty='elasticnet',
                                    solver='saga',
                                    class_weight='balanced'))
])

param_grid = {'model_LR__C': range(1,15,2),  
              'model_LR__l1_ratio': np.arange(0, 1, 0.2)}


tuning_model_LR = RandomizedSearchCV(pipeline_LR,
                                      param_grid, 
                                      cv=3, 
                                      n_jobs=-1, 
                                      scoring='roc_auc', 
                                      verbose=0,
                                      n_iter=50,
                                      random_state=140823,
                                     )

tuning_model_LR.fit(features_train, target_train)

print(f"Наилучшая метрика 'roc_auc', равная {round(tuning_model_LR.best_score_,2)},\
достигается при параметрах: {tuning_model_LR.best_params_}")

Наилучшая метрика 'roc_auc', равная 0.79,достигается при параметрах: {'model_LR__l1_ratio': 0.0, 'model_LR__C': 1}
CPU times: total: 828 ms
Wall time: 9.06 s


%%time

#  Составляем pipeline:

pipeline_RFC = Pipeline([
    ("preprocessor_ord_encoder", preprocessor_ord_encoder),
    ('model_RFC', RandomForestClassifier(random_state=140823,
                                        class_weight='balanced'))
])

param_grid = {
    'model_RFC__n_estimators': range (10, 200, 5),
    'model_RFC__max_depth': range (1,31, 2),
    'model_RFC__min_samples_leaf': range (1,11,2),
    'model_RFC__min_samples_split': range (1,21,2)}

tuning_model_RFC = RandomizedSearchCV(pipeline_RFC,
                                      param_grid, 
                                      cv=3, 
                                      n_jobs=-1, 
                                      scoring='roc_auc', 
                                      verbose=1,
                                      n_iter=100,
                                      random_state=140823,
                                     )


tuning_model_RFC.fit(features_train, target_train)

print(f"Наилучшая метрика 'roc_auc', равная {round(tuning_model_RFC.best_score_,2)},\
достигается при параметрах: {tuning_model_RFC.best_params_}")

Fitting 3 folds for each of 100 candidates, totalling 300 fits
Наилучшая метрика 'roc_auc', равная 0.88,достигается при параметрах: {'model_RFC__n_estimators': 60, 'model_RFC__min_samples_split': 9, 'model_RFC__min_samples_leaf': 1, 'model_RFC__max_depth': 11}
CPU times: total: 2.84 s
Wall time: 1min 3s


%%time

pipeline_LGBM = Pipeline([
    ("preprocessor_ord_encoder", preprocessor_ord_encoder),
    ('model_LGBM', LGBMClassifier(random_state=140823,
                                 class_weight='balanced'))
])

param_grid = {'model_LGBM__learning_rate': np.arange(0.01, 0.21, 0.05),
              'model_LGBM__num_leaves': np.arange(10, 510, 10),
              'model_LGBM__max_depth': np.arange(1, 15, 1),
              'model_LGBM__n_estimators': range (10, 1000, 10)}


tuning_model_LGBM = RandomizedSearchCV(pipeline_LGBM,
                                      param_grid, 
                                      cv=3, 
                                      n_jobs=-1, 
                                      scoring='roc_auc', 
                                      verbose=1,
                                      n_iter=100,
                                      random_state=140823
                                     )


tuning_model_LGBM.fit(features_train, target_train)

print(f"Наилучшая метрика 'roc_auc', равная {round(tuning_model_LGBM.best_score_,2)},\
достигается при параметрах: {tuning_model_LGBM.best_params_}")

Fitting 3 folds for each of 100 candidates, totalling 300 fits
Наилучшая метрика 'roc_auc', равная 0.9,достигается при параметрах: {'model_LGBM__num_leaves': 390, 'model_LGBM__n_estimators': 410, 'model_LGBM__max_depth': 2, 'model_LGBM__learning_rate': 0.16000000000000003}
CPU times: total: 2.64 s
Wall time: 1min 22s


%%time

pipeline_CBL = Pipeline([
    ("preprocessor_ord_encoder", preprocessor_ord_encoder),
    ('model_CBL', CatBoostClassifier(random_state=140823))
])

param_grid = {
    'model_CBL__depth': np.arange(1, 15, 1),
    'model_CBL__learning_rate':  np.arange(0.01, 0.21, 0.05),
    'model_CBL__iterations':  range (10, 150, 10),
    'model_CBL__l2_leaf_reg': np.arange(1, 15, 1)
}


tuning_model_CBL = RandomizedSearchCV(pipeline_CBL,
                                      param_grid, 
                                      cv=3, 
                                      n_jobs=-1, 
                                      scoring='roc_auc', 
                                      verbose=0,
                                      n_iter=30,
                                      random_state=140823
                                     )


tuning_model_CBL.fit(features_train, target_train)

print(f"Наилучшая метрика 'roc_auc', равная {round(tuning_model_CBL.best_score_,2)},\
достигается при параметрах: {tuning_model_CBL.best_params_}")

0:	learn: 0.6104719	total: 152ms	remaining: 18.1s
1:	learn: 0.5579133	total: 155ms	remaining: 9.15s
2:	learn: 0.5103059	total: 176ms	remaining: 6.88s
3:	learn: 0.4713050	total: 186ms	remaining: 5.4s
4:	learn: 0.4411582	total: 196ms	remaining: 4.5s
5:	learn: 0.4134217	total: 209ms	remaining: 3.96s
6:	learn: 0.3911923	total: 226ms	remaining: 3.65s
7:	learn: 0.3737332	total: 236ms	remaining: 3.3s
8:	learn: 0.3583523	total: 243ms	remaining: 3s
9:	learn: 0.3462995	total: 252ms	remaining: 2.77s
10:	learn: 0.3354658	total: 262ms	remaining: 2.6s
11:	learn: 0.3264000	total: 270ms	remaining: 2.43s
12:	learn: 0.3178844	total: 278ms	remaining: 2.29s
13:	learn: 0.3112632	total: 285ms	remaining: 2.16s
14:	learn: 0.3054442	total: 290ms	remaining: 2.03s
15:	learn: 0.2999026	total: 296ms	remaining: 1.93s
16:	learn: 0.2951865	total: 302ms	remaining: 1.83s
17:	learn: 0.2914498	total: 307ms	remaining: 1.74s
18:	learn: 0.2877603	total: 312ms	remaining: 1.66s
19:	learn: 0.2847123	total: 318ms	remaining: 1.59s
20:	learn: 0.2820844	total: 325ms	remaining: 1.53s
21:	learn: 0.2784046	total: 329ms	remaining: 1.47s
22:	learn: 0.2760142	total: 334ms	remaining: 1.41s
23:	learn: 0.2735383	total: 339ms	remaining: 1.36s
24:	learn: 0.2718490	total: 345ms	remaining: 1.31s
25:	learn: 0.2699109	total: 349ms	remaining: 1.26s
26:	learn: 0.2676971	total: 361ms	remaining: 1.24s
27:	learn: 0.2657717	total: 367ms	remaining: 1.21s
28:	learn: 0.2647290	total: 375ms	remaining: 1.18s
29:	learn: 0.2636886	total: 383ms	remaining: 1.15s
30:	learn: 0.2619089	total: 393ms	remaining: 1.13s
31:	learn: 0.2605894	total: 397ms	remaining: 1.09s
32:	learn: 0.2590232	total: 401ms	remaining: 1.06s
33:	learn: 0.2572242	total: 407ms	remaining: 1.03s
34:	learn: 0.2559326	total: 411ms	remaining: 998ms
35:	learn: 0.2545307	total: 415ms	remaining: 968ms
36:	learn: 0.2533173	total: 419ms	remaining: 940ms
37:	learn: 0.2519097	total: 423ms	remaining: 914ms
38:	learn: 0.2510236	total: 427ms	remaining: 888ms
39:	learn: 0.2500120	total: 432ms	remaining: 863ms
40:	learn: 0.2487785	total: 436ms	remaining: 840ms
41:	learn: 0.2471345	total: 440ms	remaining: 818ms
42:	learn: 0.2459589	total: 444ms	remaining: 796ms
43:	learn: 0.2443147	total: 449ms	remaining: 775ms
44:	learn: 0.2435919	total: 453ms	remaining: 755ms
45:	learn: 0.2425826	total: 459ms	remaining: 738ms
46:	learn: 0.2418189	total: 463ms	remaining: 719ms
47:	learn: 0.2409690	total: 467ms	remaining: 701ms
48:	learn: 0.2392035	total: 472ms	remaining: 684ms
49:	learn: 0.2385530	total: 477ms	remaining: 667ms
50:	learn: 0.2371725	total: 481ms	remaining: 651ms
51:	learn: 0.2361233	total: 486ms	remaining: 635ms
52:	learn: 0.2350281	total: 491ms	remaining: 620ms
53:	learn: 0.2342703	total: 495ms	remaining: 605ms
54:	learn: 0.2334841	total: 500ms	remaining: 591ms
55:	learn: 0.2320627	total: 505ms	remaining: 577ms
56:	learn: 0.2304878	total: 509ms	remaining: 563ms
57:	learn: 0.2296375	total: 514ms	remaining: 549ms
58:	learn: 0.2285712	total: 518ms	remaining: 536ms
59:	learn: 0.2274190	total: 523ms	remaining: 523ms
60:	learn: 0.2267351	total: 527ms	remaining: 510ms
61:	learn: 0.2257824	total: 532ms	remaining: 497ms
62:	learn: 0.2252382	total: 536ms	remaining: 485ms
63:	learn: 0.2241388	total: 540ms	remaining: 473ms
64:	learn: 0.2232916	total: 545ms	remaining: 461ms
65:	learn: 0.2225542	total: 549ms	remaining: 449ms
66:	learn: 0.2221412	total: 554ms	remaining: 438ms
67:	learn: 0.2211160	total: 562ms	remaining: 429ms
68:	learn: 0.2199636	total: 567ms	remaining: 419ms
69:	learn: 0.2189675	total: 572ms	remaining: 408ms
70:	learn: 0.2185556	total: 576ms	remaining: 397ms
71:	learn: 0.2182530	total: 581ms	remaining: 387ms
72:	learn: 0.2174748	total: 589ms	remaining: 379ms
73:	learn: 0.2170054	total: 595ms	remaining: 370ms
74:	learn: 0.2162571	total: 601ms	remaining: 360ms
75:	learn: 0.2154745	total: 605ms	remaining: 350ms
76:	learn: 0.2146147	total: 611ms	remaining: 341ms
77:	learn: 0.2141877	total: 616ms	remaining: 332ms
78:	learn: 0.2135819	total: 620ms	remaining: 322ms
79:	learn: 0.2125027	total: 625ms	remaining: 312ms
80:	learn: 0.2116462	total: 629ms	remaining: 303ms
81:	learn: 0.2109381	total: 633ms	remaining: 294ms
82:	learn: 0.2103705	total: 638ms	remaining: 284ms
83:	learn: 0.2098662	total: 643ms	remaining: 276ms
84:	learn: 0.2089565	total: 647ms	remaining: 266ms
85:	learn: 0.2080174	total: 652ms	remaining: 258ms
86:	learn: 0.2074941	total: 656ms	remaining: 249ms
87:	learn: 0.2071838	total: 661ms	remaining: 240ms
88:	learn: 0.2061283	total: 665ms	remaining: 232ms
89:	learn: 0.2050620	total: 669ms	remaining: 223ms
90:	learn: 0.2043908	total: 673ms	remaining: 214ms
91:	learn: 0.2036915	total: 677ms	remaining: 206ms
92:	learn: 0.2027861	total: 682ms	remaining: 198ms
93:	learn: 0.2023744	total: 687ms	remaining: 190ms
94:	learn: 0.2018991	total: 691ms	remaining: 182ms
95:	learn: 0.2014200	total: 695ms	remaining: 174ms
96:	learn: 0.2007151	total: 699ms	remaining: 166ms
97:	learn: 0.2001278	total: 704ms	remaining: 158ms
98:	learn: 0.1995144	total: 709ms	remaining: 150ms
99:	learn: 0.1990695	total: 713ms	remaining: 143ms
100:	learn: 0.1982690	total: 719ms	remaining: 135ms
101:	learn: 0.1980452	total: 723ms	remaining: 128ms
102:	learn: 0.1973501	total: 727ms	remaining: 120ms
103:	learn: 0.1970446	total: 731ms	remaining: 113ms
104:	learn: 0.1967605	total: 736ms	remaining: 105ms
105:	learn: 0.1963457	total: 740ms	remaining: 97.8ms
106:	learn: 0.1958531	total: 745ms	remaining: 90.5ms
107:	learn: 0.1954689	total: 749ms	remaining: 83.2ms
108:	learn: 0.1945782	total: 753ms	remaining: 76ms
109:	learn: 0.1940821	total: 758ms	remaining: 68.9ms
110:	learn: 0.1934222	total: 762ms	remaining: 61.8ms
111:	learn: 0.1927274	total: 768ms	remaining: 54.8ms
112:	learn: 0.1920778	total: 772ms	remaining: 47.8ms
113:	learn: 0.1916859	total: 777ms	remaining: 40.9ms
114:	learn: 0.1911354	total: 781ms	remaining: 34ms
115:	learn: 0.1906868	total: 788ms	remaining: 27.2ms
116:	learn: 0.1902897	total: 796ms	remaining: 20.4ms
117:	learn: 0.1896907	total: 802ms	remaining: 13.6ms
118:	learn: 0.1890179	total: 807ms	remaining: 6.78ms
119:	learn: 0.1886750	total: 811ms	remaining: 0us
Наилучшая метрика 'roc_auc', равная 0.89,достигается при параметрах: {'model_CBL__learning_rate': 0.16000000000000003, 'model_CBL__l2_leaf_reg': 2, 'model_CBL__iterations': 120, 'model_CBL__depth': 7}
CPU times: total: 2.58 s
Wall time: 5min 53s


total = [["LogisticRegression:  ", round(tuning_model_LR.best_score_,3)],
             ["RandomForestClassifier: ", round(tuning_model_RFC.best_score_,3)],
             ["LightGBMClassifier:  ", round(tuning_model_LGBM.best_score_,3)],
             ["CatBoostClassifier:  ", round(tuning_model_CBL.best_score_,3)]
            ]
total= pd.DataFrame(total, columns=["модель","ROC-AUC"])
total = total.set_index('модель')
total.index.names = [None]

total


best_model = tuning_model_LGBM


probabilities_test = best_model.predict_proba(features_test)
probabilities_one_test = probabilities_test[:, 1]

auc_roc = roc_auc_score(target_test,probabilities_one_test)

print('Метрика ROC-AUC наилучшей модели равна', round(auc_roc,2))

Метрика ROC-AUC наилучшей модели равна 0.92


# Определим долю ложноположительных ответов (FPR) и истинно положительных ответов (TPR) лучшей модели:

fpr, tpr, thresholds = roc_curve(target_test,probabilities_one_test) 

# ROC-кривая лучшей модели:
plt.figure(figsize=(8, 5))
sns.lineplot(x=fpr, y=tpr)

# ROC-кривая случайной модели: 
sns.lineplot(x=[0, 1], 
             y=[0, 1], 
             linestyle='--')\
.set(xlabel='Ложноположительный результат', 
     ylabel='Истинноположительный результат',  
     title = 'ROC-кривая')

plt.xlim(0,1) 
plt.ylim(0,1)
plt.legend(['Лучшая модель', 'Случайная модель'])
plt.show()


predictions = best_model.predict(features_test)
print('Метрика Accuracy наилучшей модели равна', round(accuracy_score(target_test, predictions),4))

Метрика Accuracy наилучшей модели равна 0.8404


cm = confusion_matrix(target_test,predictions)
cm

array([[1665,  297],
       [  74,  289]], dtype=int64)


plt.figure(figsize=(6, 6))

sns.heatmap(cm, annot=True, linewidth=.5, fmt=".0f", cbar=False)
plt.title("Матрица ошибок", fontsize=20)
plt.xlabel('Предсказания', fontsize=14)
plt.ylabel('Истинный класс', fontsize=14)
plt.show()


feature_importance = [best_model.best_estimator_[0].transformers_[0][2] + best_model.best_estimator_[0].transformers[1][2],
          list(best_model.best_estimator_[1].feature_importances_)]
         

feature_importance = pd.DataFrame(feature_importance).transpose()

feature_importance.columns =['Признаки', 'Важность']
feature_importance = feature_importance.set_index('Признаки', drop=True).sort_values(by='Важность', ascending=False)
feature_importance


plt.figure(figsize=(8, 4))

sns.barplot(feature_importance, y = feature_importance.index, x = 'Важность')
plt.title('Важность признаков', fontsize = 16)
plt.show()


# Функции для построения графиков

# Анализ дисбаланса классов
def class_imbalance(data, target):
    (data[target].value_counts()/data[target].count()*100).plot(
        kind='bar', 
        grid = True
    ).set(
        ylabel = 'Относительное количество, %', 
        xlabel = 'Факт ухода клиента: \n "No" - Клиент не ушел, "Yes" - Клиент ушел', 
        title = 'Распределение оставшихся и ушедщих клиентов' 
    )
    plt.show()

# Анализ признаков с количественными значениями
def analysis_сat_report(data, cat_col_list, cat_name_list):
    for i in range(15):
        plt.figure(figsize=(12,4))
        for b in data[col_list[i]].unique(): 
            plt.subplot(1,len(data[col_list[i]].unique()),list(data[col_list[i]].unique()).index(b)+1)
            sns.histplot(data=data[data[col_list[i]] == b], x='leave', hue ='leave',
                         shrink = 0.8, multiple='dodge', stat = 'percent', bins = 4)
            plt.title(str(i+1)+'.'+ name_list[i] + '\n- ' + b, fontsize=12)
    plt.tight_layout()

#  Анализ признаков с категориальными значениями
def analysis_num_report(data, num_list):
    for i in num_list:
        print()
        print('\033[1m' + i,':' +'\033[0m')
        print()
        sns.set()
        lines, axes = plt.subplots(1, 2, figsize=(16, 4))
        sns.histplot(data=data, hue='leave', x=i, ax=axes[0], kde=True).set_title(i + ". Плотность распределения", fontsize=18)
        sns.boxplot(data=data, x=i, y='leave', ax=axes[1]).set_title(i + ". Диаграмма размаха", fontsize=18)
        plt.show()
        
#  Создание списков для функций     
cat_col_list = ['type', 'paperless_billing', 'gender', 'senior_citizen',
            'partner', 'dependents', 'payment_method', 'internet_service', 'online_security', 'online_backup',
            'device_protection', 'tech_support', 'streaming_tv', 'streaming_movies', 'multiple_lines']
cat_name_list = ['Тип договора', 'Выставления счёта по электронной почте', 'Пол клиента', 'Пенсионный статус',
            'Наличие супруга/супруги', 'Наличие иждивенцев', 'Способ оплаты', 'Наличие услуг Интернет', 'Межсетевой экран', 
             'Облачное хранилище файлов', 'Антивирус', 'Выделенная линия технической поддержки', 
             'Онлайн-ТВ', 'Онлайн-кинотеатр', 'Возможность подключения телефонного \n аппарата к нескольким линиям']

num_list = ['monthly_charges','total_charges', 'duration', 'contract_end']


class_imbalance(df_before_delete, 'leave')


analysis_сat_report(df_before_delete, cat_col_list, cat_name_list)


analysis_num_report(df_before_delete, num_list)

monthly_charges :

total_charges :

duration :

contract_end :


plot_correlation_matrix(phik_overview.values, x_labels=phik_overview.columns, y_labels=phik_overview.index, 
                        vmin=0, vmax=1, color_map='Blues', title=r'correlation $\phi_K$', fontsize_factor=1.5,
                        figsize=(16,20))
plt.tight_layout()


total


plt.figure(figsize=(8, 5))
sns.lineplot(x=fpr, y=tpr)

sns.lineplot(x=[0, 1], 
             y=[0, 1], 
             linestyle='--')\
.set(xlabel='Ложноположительный результат', 
     ylabel='Истинноположительный результат',  
     title = 'ROC-кривая')

plt.xlim(0,1) 
plt.ylim(0,1)
plt.legend(['Лучшая модель', 'Случайная модель'])
plt.show()


plt.figure(figsize=(6, 6))
sns.heatmap(cm, annot=True, linewidth=.5, fmt=".0f", cbar=False)
plt.title("Матрица ошибок", fontsize=20)
plt.xlabel('Предсказания', fontsize=14)
plt.ylabel('Истинный класс', fontsize=14)
plt.show()


plt.figure(figsize=(8, 4))
sns.barplot(feature_importance, y = feature_importance.index, x = 'Важность')
plt.title('Важность признаков', fontsize = 16)
plt.show()

	leave	type	paperless_billing	payment_method	monthly_charges	total_charges	gender	senior_citizen	partner	dependents	internet_service	online_security	online_backup	device_protection	tech_support	streaming_tv	streaming_movies	multiple_lines	duration	contract_end
leave	1.000000	0.094015	0.083398	0.214832	0.226280	0.302890	0.008581	0.086159	0.226688	0.046871	0.056621	0.132594	0.229482	0.218380	0.103652	0.200198	0.222232	0.105101	0.374569	0.318624
duration	0.374569	0.634155	0.026799	0.350964	0.387727	0.848337	0.000000	0.063315	0.453688	0.198729	0.060845	0.395693	0.414982	0.426931	0.404795	0.339313	0.339313	0.347289	1.000000	0.730764
contract_end	0.318624	0.742251	0.184774	0.295072	0.302197	0.459746	0.000000	0.148757	0.326143	0.252733	0.262514	0.300852	0.215581	0.280061	0.336217	0.188384	0.188738	0.184148	0.730764	1.000000
total_charges	0.302890	0.470860	0.201703	0.335666	0.710905	1.000000	0.000000	0.135650	0.381958	0.084247	0.490081	0.522090	0.622445	0.640977	0.550065	0.641488	0.643210	0.467787	0.848337	0.459746
online_backup	0.229482	0.098884	0.196443	0.282475	0.629541	0.622445	0.009882	0.102065	0.219223	0.031533	0.233602	0.430425	1.000000	0.458211	0.445130	0.428007	0.417170	0.140081	0.414982	0.215581
partner	0.226688	0.179736	0.013218	0.243008	0.203545	0.381958	0.000000	0.016992	1.000000	0.652122	0.000000	0.221673	0.219223	0.238079	0.185993	0.193258	0.182011	0.086249	0.453688	0.326143
monthly_charges	0.226280	0.388444	0.467812	0.399526	1.000000	0.710905	0.008175	0.304985	0.203545	0.184366	0.919002	0.551621	0.629541	0.667481	0.576525	0.835340	0.833307	0.709983	0.387727	0.302197
streaming_movies	0.222232	0.069608	0.325551	0.378907	0.833307	0.643210	0.000000	0.186141	0.182011	0.058999	0.272782	0.289097	0.417170	0.589888	0.424078	0.742479	1.000000	0.170432	0.339313	0.188738
device_protection	0.218380	0.137610	0.160796	0.306866	0.667481	0.640977	0.000000	0.090686	0.238079	0.010416	0.232916	0.418474	0.458211	1.000000	0.499267	0.575536	0.589888	0.145710	0.426931	0.280061
payment_method	0.214832	0.277462	0.370495	1.000000	0.399526	0.335666	0.000000	0.292725	0.243008	0.224903	0.323886	0.262911	0.282475	0.306866	0.272101	0.377209	0.378907	0.174849	0.350964	0.295072
streaming_tv	0.200198	0.066961	0.343524	0.377209	0.835340	0.641488	0.000000	0.163120	0.193258	0.017331	0.272818	0.272186	0.428007	0.575536	0.422242	1.000000	0.742479	0.166899	0.339313	0.188384
online_security	0.132594	0.152145	0.000000	0.262911	0.551621	0.522090	0.018397	0.057028	0.221673	0.124945	0.241421	1.000000	0.430425	0.418474	0.528391	0.272186	0.289097	0.095572	0.395693	0.300852
multiple_lines	0.105101	0.244410	0.099953	0.174849	0.709983	0.467787	0.000000	0.087925	0.086249	0.011198	0.739808	0.095572	0.140081	0.145710	0.098571	0.166899	0.170432	1.000000	0.347289	0.184148
tech_support	0.103652	0.179999	0.055929	0.272101	0.576525	0.550065	0.000000	0.092565	0.185993	0.096912	0.239663	0.528391	0.445130	0.499267	1.000000	0.422242	0.424078	0.098571	0.404795	0.336217
type	0.094015	1.000000	0.106860	0.277462	0.388444	0.470860	0.000000	0.086231	0.179736	0.147680	0.505187	0.152145	0.098884	0.137610	0.179999	0.066961	0.069608	0.244410	0.634155	0.742251
senior_citizen	0.086159	0.086231	0.242133	0.292725	0.304985	0.135650	0.000000	1.000000	0.016992	0.324576	0.160702	0.057028	0.102065	0.090686	0.092565	0.163120	0.186141	0.087925	0.063315	0.148757
paperless_billing	0.083398	0.106860	1.000000	0.370495	0.467812	0.201703	0.000000	0.242133	0.013218	0.172593	0.231438	0.000000	0.196443	0.160796	0.055929	0.343524	0.325551	0.099953	0.026799	0.184774
internet_service	0.056621	0.505187	0.231438	0.323886	0.919002	0.490081	0.000000	0.160702	0.000000	0.108463	1.000000	0.241421	0.233602	0.232916	0.239663	0.272818	0.272782	0.739808	0.060845	0.262514
dependents	0.046871	0.147680	0.172593	0.224903	0.184366	0.084247	0.000000	0.324576	0.652122	1.000000	0.108463	0.124945	0.031533	0.010416	0.096912	0.017331	0.058999	0.011198	0.198729	0.252733
gender	0.008581	0.000000	0.000000	0.000000	0.008175	0.000000	1.000000	0.000000	0.000000	0.000000	0.000000	0.018397	0.009882	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000

	customerID	BeginDate	EndDate	Type	PaperlessBilling	PaymentMethod	MonthlyCharges	TotalCharges
0	7590-VHVEG	2020-01-01	No	Month-to-month	Yes	Electronic check	29.85	31.04
1	5575-GNVDE	2017-04-01	No	One year	No	Mailed check	56.95	2071.84
2	3668-QPYBK	2019-10-01	No	Month-to-month	Yes	Mailed check	53.85	226.17
3	7795-CFOCW	2016-05-01	No	One year	No	Bank transfer (automatic)	42.30	1960.6
4	9237-HQITU	2019-09-01	No	Month-to-month	Yes	Electronic check	70.70	353.5

	customerID	gender	Partner	Dependents
0	7590-VHVEG	Female	Yes	No
1	5575-GNVDE	Male	No	No
2	3668-QPYBK	Male	No	No
3	7795-CFOCW	Male	No	No
4	9237-HQITU	Female	No	No

	begin_date	end_date	leave
9	2014-12-01	2017-05-01	Yes
15	2014-05-01	2016-03-01	Yes
25	2017-08-01	2018-09-01	Yes
30	2014-03-01	2018-11-01	Yes
35	2014-02-01	2018-12-01	Yes

	customer_id	begin_date	end_date	type	paperless_billing	payment_method	monthly_charges	total_charges	gender	senior_citizen	partner	dependents	internet_service	online_security	online_backup	device_protection	tech_support	streaming_tv	streaming_movies	multiple_lines	leave
936	5709-LVOEQ	2020-02-01	NaT	Two year	No	Mailed check	80.85	NaN	Female	No	Yes	Yes	DSL	Yes	Yes	Yes	No	Yes	Yes	No	No
1340	1371-DWPAZ	2020-02-01	NaT	Two year	No	Credit card (automatic)	56.05	NaN	Female	No	Yes	Yes	DSL	Yes	Yes	Yes	Yes	Yes	No	NaN	No
753	3115-CZMZD	2020-02-01	NaT	Two year	No	Mailed check	20.25	NaN	Male	No	No	Yes	NaN	NaN	NaN	NaN	NaN	NaN	NaN	No	No
4380	2520-SGTTA	2020-02-01	NaT	Two year	No	Mailed check	20.00	NaN	Female	No	Yes	Yes	NaN	NaN	NaN	NaN	NaN	NaN	NaN	No	No
1082	4367-NUYAO	2020-02-01	NaT	Two year	No	Mailed check	25.75	NaN	Male	No	Yes	Yes	NaN	NaN	NaN	NaN	NaN	NaN	NaN	Yes	No

	customer_id	begin_date	end_date	type	paperless_billing	payment_method	monthly_charges	total_charges	gender	senior_citizen	partner	dependents	internet_service	online_security	online_backup	device_protection	tech_support	streaming_tv	streaming_movies	multiple_lines	leave
2689	8628-MFKAX	2014-02-01	2020-02-01	Two year	Yes	Credit card (automatic)	116.75	8910.36	Female	Yes	Yes	No	Fiber optic	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No
2515	8869-LIHMK	2014-10-01	2020-02-01	Two year	Yes	Bank transfer (automatic)	115.10	8029.38	Female	No	No	No	Fiber optic	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No
5995	2193-SFWQW	2014-02-01	2020-02-01	Two year	No	Bank transfer (automatic)	111.95	8060.40	Male	No	Yes	Yes	Fiber optic	Yes	Yes	Yes	Yes	Yes	Yes	No	No
3348	2172-EJXVF	2014-03-01	2020-02-01	One year	Yes	Electronic check	105.90	7744.47	Female	Yes	No	No	Fiber optic	No	Yes	Yes	No	Yes	Yes	Yes	No
1280	2388-LAESQ	2014-02-01	2020-02-01	Two year	Yes	Bank transfer (automatic)	114.85	8765.35	Female	Yes	Yes	No	Fiber optic	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No
6035	9835-ZIITK	2014-06-01	2020-01-01	One year	Yes	Electronic check	110.85	7649.76	Male	Yes	Yes	No	Fiber optic	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes
57	5067-XJQFU	2014-08-01	2020-02-01	One year	Yes	Electronic check	108.45	7730.32	Male	Yes	Yes	Yes	Fiber optic	No	Yes	Yes	Yes	Yes	Yes	Yes	No
3964	2632-IVXVF	2014-06-01	2020-02-01	Two year	No	Credit card (automatic)	111.75	7902.96	Female	No	Yes	Yes	Fiber optic	Yes	No	Yes	Yes	Yes	Yes	Yes	No
6225	1452-UZOSF	2014-02-01	2020-02-01	Two year	Yes	Credit card (automatic)	106.10	7639.20	Male	No	Yes	Yes	Fiber optic	Yes	Yes	Yes	Yes	Yes	No	Yes	No
4031	8309-PPCED	2014-02-01	2020-02-01	Two year	Yes	Bank transfer (automatic)	110.45	8429.54	Female	No	Yes	No	Fiber optic	No	Yes	Yes	Yes	Yes	Yes	Yes	No

	ROC-AUC
LogisticRegression:	0.790
RandomForestClassifier:	0.875
LightGBMClassifier:	0.897
CatBoostClassifier:	0.894

	Важность
Признаки
contract_end	423
duration	301
total_charges	172
monthly_charges	161
payment_method	38
type	32
partner	22
online_backup	16
tech_support	13
multiple_lines	13
senior_citizen	12
device_protection	9
paperless_billing	7
streaming_movies	6
online_security	3
streaming_tv	1

Содержание

Модель машинного обучения, предсказывающая уход клиентов оператора связи «Ниединогоразрыва.ком»¶

План работы¶

Цель и задачи работы¶

Первоначальное ознакомление с данными¶

Объединение данных¶

Изменение названия столбцов¶

Выделение целевого признака¶

Исследование данных¶

Предварительный аналих данных¶

Изменение типа данных¶

senior_citizen¶

total_charges¶

begin_date, end_date¶

Заполнение пропусков¶

total_charges¶

end_date¶

multiple_lines¶

internet_service, online_security, online_backup, device_protection, tech_support, streaming_tv, streaming_movies¶

Анализ признаков с количественными значениями¶

Анализ признаков с категориальными значениями¶

Удаление и создание признаков¶

Создание признаков¶

Анализ синтетических признаков¶

Удаление избыточных признаков¶

Корреляционный анализ¶

Проверка корреляции признаков¶

Удаление признаков c низкой корреляцией и мультиколлиниарностью¶

Выводы по 1 разделу¶

Разбиение датасета на обучающую и тестовую выборки¶

Кодирование и масштабирование признаков¶

Выбор и обучение моделей машинного обучения на кросс-валидации¶

LogisticRegression¶

RandomForestClassifier¶

LGBMClassifier¶

CatBoostClassifier¶

Выбор лучшей модели по метрике ROC-AUC на кросс-валидации¶

Оценка качества лучшей модели на тестовой выборке¶

Построение и анализ ROC-кривой, accuracy, матрицы ошибок и важности признаков лучшей модели на тестовой выборке.¶

ROC-кривая¶

Accuracy¶

Матрица ошибок¶

Важность признаков¶

Выводы по 2 разделу¶

Отчет о проделанной работе¶

Выполнение плана¶

Трудности, возникшие при выполнении работы:¶

Ключевые шаги в решении задачи¶

Подготовка данных¶

Обучение моделей машинного обучения¶

Проверка качества выбранной модели¶

Рекомендации о введении модели в эксплуатацию (или её переработки)¶